Use cases by function9 use cases

Operations automation for the coordination work that eats the day

Operations work is mostly moving information between people and systems. An order in a PDF has to become a sales order. A supplier's reply changes a delivery date that three other people depend on. A tenant's photo of a leak has to become a work order and a plumber's visit. A grid operator's form needs the same customer data typed for the fifth time this week. None of it needs judgment most of the time, and all of it is too irregular for classic rule-based automation.

That combination, repetitive but irregular, is exactly where language models help, provided the systems of record stay in charge. It is also where projects fail for reasons that have nothing to do with the model: Anthropic's 2026 survey of companies building agents found integration with existing systems to be the top barrier, cited by 46%. The ERP's API, the portal that has none, and the exceptions are the real work.

The builds here include customer orders entered into the ERP from email and WhatsApp, a shared inbox that sorts and files itself, suppliers chased for delivery dates, maintenance requests with a real emergency path and grid-connection paperwork filed by an agent.

Operations and logistics systems I can build

How these builds stay safe inside real operations

  • The ERP stays the source of truth

    Stock, prices, customers and orders are read from the system of record at the moment they are needed, and results are written back through its API, not kept in a parallel spreadsheet.

  • Every write is idempotent

    An order, a work order or a status update is created once, however many times a message is retried or forwarded. That single rule prevents most of the incidents that make teams distrust automation.

  • Uncertainty is handled line by line

    A purchase order with twenty lines and one unclear part number should not stop the other nineteen. Low-confidence lines go to a person; the rest flow through.

  • No API is not a dead end

    Portals and old desktop systems can be operated by a supervised browser or computer-use agent, with a person approving the final submission and every step recorded.

  • Exceptions are counted

    The volume and type of exceptions is tracked weekly. It shows where the process itself is broken, which is often more valuable than the automation.

Frequently asked questions

Can AI enter orders into SAP Business One or NetSuite?

Yes. SAP Business One exposes the Service Layer, NetSuite has SuiteTalk and RESTlets, and both can create sales orders through their APIs. The work is in mapping customer part numbers to your items, checking prices and stock, and deciding what happens to lines the model is unsure about, which go to a person.

How do I automate order processing from email?

Capture every order channel into one queue, extract the lines with a model under a strict schema, match them against your items and contract prices in code, create the order through the ERP's API, and send the confirmation. Start in draft mode, where a person approves each order, and switch to automatic for customers whose orders consistently match.

What if the system we use has no API?

Then the options are a database connection if one is permitted, a file export and import, or a browser or computer-use agent that operates the screens the way a person would. The last option is slower and needs a person to approve submissions, but it is now reliable enough for repetitive portal work.

Will this replace our operations team?

It replaces the typing and the chasing, not the team. The people who knew which supplier always ships late and which customer's orders need checking become the ones who handle exceptions and improve the rules. In most operations teams that work was being squeezed out by data entry anyway.

Also browse: Manufacturing and distribution · Logistics and freight · Outlook and Microsoft 365 · SAP Business One · all use cases